Learning from traffic data collected before, during and after a hurricane

(1)

Learning from traf

ﬁc data collected before, during and after a hurricane

Erik Archibald

a

_{, Sue McNeil}

b,

⁎

a

Disaster Research Center, University of Delaware, Newark DE 19716, USA b

Department of Civil and Environmental Engineering, University of Delaware, Newark DE 19716, USA

a b s t r a c t

a r t i c l e i n f o

Article history: Received 7 April 2012

Received in revised form 28 June 2012 Accepted 30 June 2012 Keywords: Disasters Evacuation Trafﬁc counts Hurricane Trafﬁc patterns Emergency services

Hurricanes harm people and damage property through extreme wind speeds andflooding associated with heavy rains and storm surge. One of the most effective and widely used tactics to protect people from hurri-canes is evacuation. Improved knowledge of the behavior of communities before, during and after an evacu-ation can better support emergency planning and operevacu-ations, and thus help make evacuevacu-ations safer and more efficient. The objective of this work is to identify ways to use traffic data to better understand evacuation be-havior and to explore ways to integrate traffic data into evacuation planning and response. Traffic data col-lected in Delaware before, during and after Hurricane Irene in August 2011 using automated traffic recorders are assembled and analyzed. The analysis shows that a significant number of residents and visitors evacuated from the beach communities and the evacuation patterns are very similar to the traffic patterns experienced on summer weekends. These insights suggest that this type of analysis may also be of value for other events in other communities.

1. Introduction 1.1. Motivation

Experiences over the past two decades provide clear evidence of the destructive power of hurricanes, also known as cyclones and ty-phoons. These experiences demonstrate the importance of develop-ing strategies to prepare for and mitigate the impacts of hurricanes. Hurricanes harm people and damage property through extreme wind speeds and ﬂooding associated with heavy rains and storm surge. Hurricanes are a major problem in the U.S. and around the world. Hurricanes cause about $10 billion in damages in the U.S. each year[1]. Beyond recurring yearly losses, hurricanes also have the potential to cause catastrophic loss. For example, in 1970, a ty-phoon caused one of the largest catastrophes in world history, killing over 300,000 people in Bangladesh. More recently, in 2005, Hurricane Katrina caused $108 billion in damage and killed 1200 people on the Gulf Coast of the U.S[2]. Hurricanes have been and will continue to be a cause for concern as population centers in coastal areas around the world continue to grow[3].

One of the most effective and widely used tactics to protect people from hurricanes is evacuation[3]. Evacuation allows people to escape the hazards of extreme wind andﬂooding. In an evacuation, residents and visitors of hazardous areas leave before they are exposed to neg-ative consequences. Evacuations are especially feasible for hurricanes as advance warnings allow large populations to leave at risk areas

be-fore the hurricane arrives. While evacuations hold the potential to protect the public, they may put evacuees at unnecessary risk, and certainly cost government, industry and private citizens' time and money. Improved knowledge of the behavior of communities before, during and after an evacuation can better support emergency plan-ning and operations, and thus help make evacuations safer and more efﬁcient.

As lead time in issuing hurricane warnings to threatened populations increases with improvements in hurricane monitoring and mass communication technology, the social and organizational features of integrated warning systems become paramount as key factors in saving lives and reducing damages to property. Even if the public understands hurricane forecasts, their trust in the reli-ability and accuracy of these forecasts, and in the sources that pro-vide such information, may significantly impact their behavior and response[4,5]. For example, public con_{fidence and trust in the} sources that provide such information (e.g., hurricane forecasts and warnings) has an impact on their perception of risk[6,7]. However, trust in institutions is a variable entity, often a function of minority sta-tus and power[5]that at times is undermined by mass media accounts that convey inaccurate, biased, and exaggerated information[8–11]. Sufficient lead time, moreover, should allow the public to take appropri-ate action. Previous research has shown that one of the most significant problems with weather forecasts is how the information is presented and communicated to end-user communities (e.g., government agen-cies, emergency management organizations, industry, and to the gener-al population; see Refs.[4,12–14]). It is noteworthy, however, that even forecasts of severe weather events that attempt to solve these problems may fail to elicit appropriate protective action (horizontal evacuation,

⁎ Corresponding author. Tel.: +1 302 831 6578.

E-mail addresses:[email protected](E. Archibald),[email protected](S. McNeil).

Contents lists available atSciVerse ScienceDirect

(2)

vertical evacuation, shuttering, securing roofs, stockpiling food, having a generator, alternate means of communicating etc.) given that an individual's response to forecasts and warnings is often impacted by factors that have little to do with the technical features of weather fore-casts, such as the individual's social class, education, gender, race, eth-nicity, cultural background, and previous experiences with weather events.

Given the different perceptions of events, access to and response to information, and alternative actions, we know very little about who actually evacuates and when, if a hurricane warning is issued, and when the evacuees return. This lack of information is a problem for both planning for and responding to hurricanes.

1.2. Objective

Traditionally, emergency management agencies perform the majority of evacuation planning and operations, but recently trans-portation departments are becoming increasingly involved[3]. This provides an opportunity for new improvements to evacuation planning and operations. Specifically, increased participation of Departments of Transportation (DOTs) benefits evacuation opera-tions by involving transportation professionals and utilizing intelli-gent transportation systems (ITS). Transportation professionals and ITS provide access to real-time collection and analysis of traffic data that can be used to inform the planning process and enhance the efficiency of response operations.

The objective of this work is to identify ways to use traffic data to better understand evacuation behavior and to explore ways to inte-grate traf_{fic data into evacuation planning and response. This involves} examining traffic data collected in Delaware before, during and after Hurricane Irene in August 2011. Analysis of this data is performed and the usefulness of this type of data is then discussed. The use of traffic data will ultimately allow governments to better plan and exe-cute evacuations.

1.3. Related work

While a great deal of research focuses on evacuations in general and hurricane evacuations in particular[15–17], relatively few studies make use of traffic data. Many evacuation studies instead rely on surveys. For example, Kang et al.[18]compared hypothesized household behavior with actual behavior during a hurricane. Other studies have used traf_fic data to support their data collected from phone interviews. For exam-ple, Dow and Cutter reviewed data from Hurricane Floyd[15]. A notable use of traf_{fic data is the analysis of the 2005 Hurricane Katrina} evacua-tion in Louisiana by Wolshon[17,19,20]. Traffic data analyses have also been undertaken for other types of disasters. In Japan, traffic data pro-vided insight into the impact and recovery from the 1995 Kobe earth-quake[21]. Other studies use travel demand and traffic models to plan evacuation routes[22,23]and some recognition has been given to the importance of integrating behavioral information with the models[24].

Among state DOTs monitoring and usage of traffic data are far more common. Across the country, states have been working on de-veloping and implementing ITS systems using inductive loop traffic counters and closed circuit television to monitor real-time travel con-ditions[25]. In hurricane-prone areas, state plans identify these tech-nologies as an important resource[26]. Florida has a statewide web based system to monitor traffic data for hurricane response and parts of Texas use real-time data for hurricane response as well[27]. In Delaware, the area examined in this study, the use of real-time data for hurricane evacuation is not standard practice as it is in states that regularly experience hurricanes. Delaware like other states with some hurricane risk does, however, have planning documents to ad-dress hurricane hazards. In Delaware, consultants have worked with Delaware Department of Transportation and Delaware Emergency

Management Agency to develop all hazards evacuation plans[28–30]. The Army Corps of Engineers has also developed a Hurricane Evacuation Study to inform hurricane evacuation planning for the entire Delmarva Peninsula, which includes Delaware as well as parts of Maryland and Virginia. This study also deﬁnes behavioral assumptions for the Delmarva Peninsula[31].

2. Data and data analysis 2.1. Context for the data

Delaware is an Atlantic coastal state. While Delaware is the sec-ond smallest state (under 2000 square miles or about 5000 square kilometers) and has a population of about 900,000, vacationers swell the population of the beach communities each summer. The state is situated on the Delmarva Peninsula, a narrow body of land bordered by Chesapeake Bay to the west and Delaware Bay and the Atlantic Ocean to the east. The state of Delaware is bordered by Pennsylvania to the north and Maryland to the west and south. The Chesapeake–Delaware canal crosses northern Delaware providing a shorter shipping route between Baltimore and Philadelphia. The ma-jority of the population is clustered in northern Delaware away from the ocean. State Route 1 and US 13 are the primary north–south routes. State Route 1 is a multi-lane limited access toll road in north-ern Delaware. Along the Atlantic Ocean, the beach communities are developed with a dense local street network. Most access is from the north with limited access to the west, and to the south via the Chesapeake Bay Bridge and Tunnel. This geography means that there is limited access to the beach communities. During severe weather, this may be compounded by closures at the bridges across the canal, and the Indian River Inlet in southern Delaware.

Although the state frequently experiences severe weather, tropi-cal storms and hurricanes are less common in the area. In theﬁve year period between 2005 and 2010, Delaware experienced 37 se-vere weather events, of which only one was a tropical storm[32]. Hurricane Irene in August 2011 was theﬁrst time Delaware experi-enced a hurricane evacuation.

This work uses data from Hurricane Irene to explore the useful-ness of traffic data for modeling and understanding hurricane evacu-ations. Traffic data from Delaware before, during and after the 2011 Hurricane Irene are analyzed, and graphed to identify traffic patterns before, during and after the mandatory evacuation. This information is used to make some inference about the behavior of the population. To put this analysis further in context, the timeline of the event is reviewed.

On Thursday August 25, 2011, the National Weather Service is-sued a hurricane warning for coastal Delaware[33] and an hour later the governor issued an order for the immediate evacuation of all visitors in coastal areas effective at 6:00 PM, August 25th, local time. This evacuation order was amended Friday August 26th at 11:00 AM to include a mandatory evacuation for all residents within three-quarters of a mile (1.1 km) of major bodies of water. Delaware coastal residents had until 9:00 AM Saturday to evacuate[34]. In nearby Maryland, the timing and conditions of evacuations orders varied widely, as individual counties and municipalities made the de-termination of what kind of evacuation order to issue and when. Ocean City, Maryland, a coastal community bordering Delaware, pur-sued one of the most aggressive evacuations in the state, calling for a mandatory evacuation of all residents and visitors, requiring busi-nesses to close, banning the sale of alcohol beginning Friday morning at midnight, restricting entry to only essential personnel and going door to door Friday to make sure everyone had left[35]. Irene was a category 1 hurricane as it passed just off the coast of the Delmarva Peninsula between 8 PM Saturday night, and 5 AM Sunday morning[33]. August is a busy time for the beach communities in southern Del-aware. In addition to the residents there are the regular beachgoers

(3)

and visitors. Regular beachgoers live in the surrounding region and own property at the beach or have a rental for the season. Visitors may be vacationers staying in motels, campgrounds or rental proper-ties or day visitors coming from the surrounding area. Short term beach rentals typically rent from Saturday to Saturday. Occupancy rates are typically 100% in August. The weekend following the hurri-cane was Labor Day weekend, which is traditionally a busy holiday weekend.

2.2. Data sources

The primary sources of data for this project are traffic counts from the Delaware Department of Transportation (DelDOT). These traffic counts are collected using inductive loop detectors. DelDOT maintains 75 permanent automatic traffic recorders which provide continuous volume data in 15 minute or 1 hour increments[36]. In addition to per-manent dedicated counters, DelDOT provides count data from many of the traffic signals in operation across the state. Data are not continuous-ly available for all devices due to loss of solar power or other issues.

For this analysis, data from a number of selected sites are examined to see the impact of hurricane evacuation. The sites were selected to provide insights into travel patterns from areas under mandatory evacuation. Population, housing and demographic data for the analysis are extracted from United States 2010 Census data using ESRI Community Analyst.

The trafﬁc counters selected are as follows:

• Four count locations are used to look at evacuation behavior from a 10.6 square mile (27.5 square kilometer) area that includes the small beach communities of Bethany Beach, South Bethany and Fenwick Island. A trafﬁc counter on DE 1 just north of the Indian River Inlet Bridge provides data on northward evacuation, counters on DE 26 and 54 capture the majority of the westward movement and a counter along DE 1 provides information about movement into and out of Maryland. Together these counters serve as a cordon and capture almost all of the trafﬁc entering and exiting the Betha-ny Beach and Fenwick Island area.

• A trafﬁc counter is located on DE 1 just south of DE 16. This counter provides information about the number of vehicles moving south into and north out of a larger area of beaches that contains approx-imately 96 square miles (248.6 square kilometers) of Rehoboth Beach, Dewey Beach and Lewes in addition to the small communi-ties further south.

These locations are shown inFig. 1. Data were acquired for the pe-riod August 12 through to September 19, 2011. This pepe-riod provides a picture of weekly and daily trafﬁc patterns before and after the hurricane.

3. Data analysis, theory and calculations

To explore and analyze the traffic count data, and infer informa-tion about evacuainforma-tion behavior, graphical analysis, summary statis-tics, and principles offlow conservation are used. Demographic data and traffic count data can be used to estimate the number of people moving in the region. This information reflects disruptions to travel patterns and may be useful to better inform evacuation planning and operations.

3.1. Graphical analysis of trafﬁc ﬂow data

Traffic data typically show regular peaking characteristics that reflect seasonal, day of the week and time of day variations and are distorted by events. Graphing traffic against time provides insights into how much variation occurs in these patterns and helps to identify disruptions.

3.2. Using conservation ofﬂow to estimate vehicles evacuated

If suf_{ficient data are available, traffic counts can be used to} esti-mate the population evacuated. This can be done by doing a simple mass balance, except using number of vehicles instead of mass. Auto-matic traf_{fic recorders count the number of vehicles passing a given} point over a specified period of time. This period of time might be a single day or a number of days over which the evacuation occurred. If enough counters are available to capture all traffic entering and exiting the area, one can pinpoint exactly how many vehicles left a given area in the specified period. Eq.(1)describes how the change in vehicle count for an area can be calculated.

Δn ¼ ∑AllcountersðInbound Count−Outbound CountÞ ð1Þ

A positiveΔn indicates cars entering the area, and a negative value means that they are leaving.

3.3. Measuring population and percentage evacuated

With a little more information, one can estimate how many people evacuated in this time period. This is possible by multiply-ing by the vehicle occupancy rate or number of people per vehicle. This value can either be assumed or measured. Since entire house-holds are evacuating it is possible that the vehicle occupancy is not the same as normal conditions. In Delaware, hurricane planning as-sumptions do not give guidance on the number of people per vehi-cle, but they do provide the number of vehicles per household that would be used in an evacuation and the number of people per household is found in U.S. Census data. Eq.(2)is used toﬁnd the number of evacuees using the number of vehicles that evacuated (Δn), assuming the number of vehicles per household is the same for the residents and the visitors.

Number of Evacuees¼ Δnj j people per householdð Þ = vehicles per householdð Þ

ð2Þ Once this is determined, GIS (geographic information systems) and demographic data can assist in estimating the percentage of the population that evacuated. Census data provide the number of per-manent residents and also the number of seasonal rental units in that area. Other data are needed or assumptions must be made to de-termine how many tourists are really in the area. This can be done by identifying the percent occupancy of seasonal units and the number of people in each as seen in Eq.(3).

Total Population¼ permanent residents þ % occupancy

seasonal units visitors per unit ð3Þ Once the total population is known, the percentage evacuated can be found by dividing the number of evacuees by the total population (% Evacuated = Number of evacuees/Total Population).

Trafﬁc data can also provide useful information about the usage of roadway capacity for evacuation. This can be done by simply dividing the hourly volume by the capacity for each segment of roadway. 4. Results

4.1. Hourly, daily and weekly trafﬁc ﬂow patterns

To better understand the traffic patterns this analysis focuses on the traffic data from Route 1 south of the Indian River Inlet Bridge. Alternate routes to Route 1 are inland and this location provides a good indication of temporal traffic patterns for northbound (NB) and southbound (SB) traffic from and to the beach communities of Bethany Beach, South Bethany and Fenwick Island in southern Delaware.Fig. 2shows graphs

(4)

by week (Monday to Sunday) for NB and SB hourly trafﬁc counts for State Route 1 S of the Indian River Inlet Bridge for the week prior to, the week of and the week following the hurricane. The graphs are par-ticularly useful to communicate the diurnal and weekly trafﬁc patterns visually.

NB traffic typically shows a morning and afternoon peak where SB traffic generally shows an afternoon and a smaller evening peak. The anticipated impacts of the approaching hurricane are reflected in the traffic flow data. The day the evacuation order was issued, August 25, NB traffic shows a more extended morning peak and a higher after-noon peak. On August 26, the NB traffic flow showed a single peak and SB peak traffic flow was approximately 25% of the peak value the previous week. On August 27, the day the storm arrived, there was almost no traffic in both the NB and SB directions. The Indian River Inlet Bridge was closed to traf_{fic 4 PM Saturday August 27 and} did not reopen until after 4 PM Sunday August 28 when engineers had checked for scour. On Monday August 29, SB traffic was much higher than NB but lower than usual. Traf_{fic is typically heavier on} Fridays, Saturdays, Sundays and Mondays than Tuesdays, Wednes-days and ThursWednes-days.

Table 1presents another way to look at these data. This shows the directional volume for each day, the difference and the total volume. The difference indicates whether vehicles have left the area (nega-tive) or are entering the area (posi(nega-tive). The total volume (sum) is a good indication of activity or transportation demand. In this case, we see that activity is typically very high on weekends, but is very low the weekend of the hurricane. Activity was also lower the week immediately after the hurricane. Even though the weather was sunny and nice, this week is near the end of beach season.

In a typical summer week, peopleﬁnishing weekend trips head back north on Sunday and Monday, and the area begins to accumulate more people and vehicles on Tuesday and Wednesday. On Thursday, Friday and Saturday large masses come down to begin vacation or stay for the weekend. The week before the week of the hurricane, about the same number of people entered and left the area (190 more vehicles traveled SB than NB in the week of August 14 to August 20). The week of the hurricane (Sunday August 21 to Saturday August 27), 13,215 more vehicles traveled north than south. In the week of the hurricane, the Sunday and Monday before the hurricane were similar to the previous week. However unlike other weeks in which

(5)

people begin to trickle in on Tuesday and Wednesday, the Tuesday and Wednesday before the hurricane people also showed a net loss. This is probably not because they were evacuating due to the hurri-cane. It is more likely that there was a net loss on Tuesday and Wednesday because as peopleﬁnished their vacations and headed home, those who would regularly come to begin vacations canceled their plans. On Thursday the governor ordered a mandatory evacua-tion of all non-residents in coastal areas. This was probably a large factor in the net loss of 3256 vehicles on Thursday. Residents were given the deadline to evacuate by 9 AM Saturday morning. On Friday 5731 vehicles left the area and on Saturday a few more left as well.

Of the 13,215 vehicles that left the area, it is important to consider how many of these were permanent residents and how many were visitors. The pattern in which residents returned helps answer this question. Residents would probably be more likely to return to their homes quickly while vacationers would probably just cancel their va-cations or go elsewhere. Traffic patterns indicate that most visitors do not arrive until the weekend. Using the data fromTable 1, the cumu-late net_{flow NB can be plotted over the period from August 14 to} September 10 as shown inFig. 3. Assuming the transportation net-work functions in a similar manner during the evacuation and normal peak periods (for example, Sundays), then the difference between the peak cumulative net NBflow on August 27 and the peak cumulative

net NBﬂow on September 6 is 4566 vehicles. This net ﬂow provides an indication of the number of residents evacuating via this route. 4.2. Peaking patterns

Many times planners have a hard time understanding an emer-gency scenario such as a hurricane evacuation because these events do not happen very often. Examining trafﬁc data after a hurricane evacuation can help planners better understand what conditions are like during a hurricane.

Analysis of traffic data can help planners make comparisons which allow them to understand evacuation traffic better. In the case for the beach areas of Delaware, traffic operations personnel recognize the sim-ilarity of hurricane evacuation with normal beach traffic. In the words of DelDOT's Traffic Management Center Director, Gene Donaldson:

“Moving beach trafﬁc is a familiar task for DelDOT, which does so every summer weekend… We understand the road network, how much traf_{ﬁc we can move. We work closely with Maryland”}[37]

Traffic data from Hurricane Irene confirm the similarity of normal beach traffic and hurricane evacuation traffic. In this case a compari-son of traffic data from before the hurricane (Sunday, August 14),

8/15/11 8/16/11 8/17/11 8/18/11 8/19/11 8/20/11 8/21/11 8/22/11

Vehicles Per Hour

1000 1200 200 400 600 800 0

1000 1200 200 400 600 800 0 8/22/11 8/23/11 8/24/11 8/25/11 8/26/11 8/27/11 8/28/11 8/29/11 8/29/11 8/30/11 8/31/11 9/1/11 9/2/11 9/3/11 9/4/11 9/5/11

SR 1 @ Indian River SB SR 1 @ Indian River NB

(6)

the day of the evacuation (Thursday, August 25) and Labor Day (Mon-day, September 5) inFig. 3helps people see that evacuation trafﬁc might not be totally unexpected. In an area where the number of tourists in coastal areas far exceeds the permanent population, it is reasonable that a day when all the tourists leave would be quite sim-ilar to an evacuation in which everyone closest to the coast leaves.

Fig. 4shows the similarity between evacuation traffic, Labor Day traf-fic and Sunday traftraf-fic at two different points along Coastal Highway, Route 1.

The graphs (Fig. 4) show that peak evacuation traffic actually ended up being less than both Labor Day traffic and Sunday traffic from the 2 weeks prior. This knowledge helps transportation and emergency management officials understand what they are dealing with and respond appropriately.

4.3. Estimating the population evacuation

The methods fromSection 3are used to identify how many evac-uees left the coastal areas of Bethany Beach, South Bethany and Fenwick Island. This is done using the trafﬁc counters at the Indian River Inlet Bridge on DE 1, DE26 at Assawoman Bridge, DE 54 at Har-poon Hannahs and DE 1 at the Maryland border. These counters serve as a cordon around the area and capture most, but not quite all trafﬁc entering and exiting a 10.6 square mile area. According to the ESRI Community Analyst's processing of the 2010 census, this area is home to about 4372 permanent residents living in 2271 households.

Beyond the resident population, in the summer the area contains many tourists and has about 8944 housing units available for season-al/recreational use recorded in the 2010 census. The trafﬁc count lo-cations and area are shown inFig. 5.

We can calculate how many people evacuated from this area for a given time period using the trafﬁc counts. From the time of the ﬁrst evacuation order (8/25 18:00) to the evacuation deadline (8/27 9:00) 24,867 vehicles entered the area and 35,305 exited the area using four different routes. By subtracting the two we see that net 10,438 vehicles left the area during the period of the evacuation. Most of these evacuated by heading north on DE 1 across the Indian

Table 1

Daily trafﬁc volume data for DE 1 at Indian River.

Day Date SB NB Difference

(SB-NB)

Sum

(NB+SB) Interpretation (for route 1 coastal hwy)

Governor orders immediate evacuation of non-residents Largest day for evacuation.

Most evacuees leave by 9:00 AM deadline Day after storm, many people come back. Monday, lots of people come back.

Sun 8/14/2011 9479 12276 Mon 8/15/2011 11154 11598 Tue 8/16/2011 11216 11114 Wed 8/17/2011 11622 11142 Thu 8/18/2011 12722 11929 Fri 8/19/2011 14086 12707 Sat 8/20/2011 13918 13241 Sun 8/21/2011 9507 12210 Mon 8/22/2011 10825 11728 Tue 8/23/2011 9890 10200 Wed 8/24/2011 10606 10614 Thu 8/25/2011 10232 13488 Fri 8/26/2011 3415 9146 Sat 8/27/2011 340 644 Sun 8/28/2011 2095 1134 Mon 8/29/2011 8262 4810 Tue 8/30/2011 8020 6679 Wed 8/31/2011 8700 7729 Thu 9/1/2011 11109 8429 Fri 9/2/2011 14570 10104 Sat 9/3/2011 14167 12569 Sun 9/4/2011 11521 13732 Mon 9/5/2011 6486 13228 Tue 9/6/2011 7826 9776 Wed 9/7/2011 6642 6625 Thu 9/8/2011 7462 6950 Fri 9/9/2011 10380 7974 Sat 9/10/2011 9311 8617 -2797 21755 -444 22752 102 22330 480 22764 793 24651 1379 26793 677 27159 -2703 21717 -903 22553 -310 20090 -8 21220 -3256 23720 -5731 12561 -304 984 961 3229 3452 13072 1341 14699 971 16429 2680 19538 4466 24674 1598 26736 -2211 25253 -6742 19714 -1950 17602 17 13267 512 14412 2406 18354 694 17928

Early birds start to make it to beach for vacation

Most people come to beach in time for weekend

People head inland back to work.

Most people come to beach in time for weekend People head inland back to work.

Early birds start to make it to beach for vacation

Most people come to beach in time for weekend

People head inland back to work.

People have heard of hurricane, cancel weekend beach plans

2000 4000 6000 8000 10000 12000 14000 -4000 -2000 0 8/14/2011 8/19/2011 8/24/2011 8/29/2011 9/3/2011 9/8/2011

Cumulative Net NB Flow

(7)

River bridge or by heading west on DE 26. By making an assumption about the occupancy of each vehicle, we can estimate how many peo-ple evacuated in this time period. In 2006 a study on evacuation be-havior in the Delmarva Peninsula found that on average people would use 1.3 cars per household to evacuate[31]. U.S. 2010 census data put together for the area indicate that there are 1.93 people per household. This information can be used toﬁnd how many people evacuated using this route, in this time window using Eq.(2).

Using Eq.(2)reveals that approximately 15,461 people (10,438 vehicles times 1.93 people per household divided by 1.3 vehicles per household) left the area during the time period in which the evac-uation order was in effect (8/25 18:00 to 8/27 9:00). This same meth-od was used to determine that from noon the Saturday before the hurricane to noon the Saturday of the hurricane there were 44,693 fewer people. This includes not only those who evacuated directly be-fore the hurricane, but also accounts for the fact that over the course of the week before the hurricane many visitors ended their vacations as planned and went home while new visitors did not come in to re-place them.

ESRI Community Analyst was used to estimate population, house-holds and number of vacation units for the speciﬁed area. This anal-ysis shows that the area has 4372 people living in 2271 households and has 8944 seasonal residential use units. In addition, searches of websites for campgrounds, motels, hotels and bed and breakfasts identiﬁed another 1493 units. Using an assumption of 90% for occu-pancy rate for the seasonal housing and 2.05 visitors per unit we can estimate the total population [31]. This estimates the total

population as being 23,628 people. Given that this is peak season, as-suming 100% occupancy, then the population is estimated to be 25,769.

The percent of the residents and visitors that evacuated can be found by dividing the total number of evacuees by the total popula-tion. In this case the percent evacuated is approximately 60_{–65%, a} signiﬁcant portion of residents and visitors.

5. Discussion

5.1. Real-time trafﬁc data applications for operations

In a disaster, emergency operations personnel continually seek new information to understand how an incident is developing and how their actions are making an impact. This information builds a sit-uational awareness of the circumstances they face[27]. Increased sit-uational awareness helps officials monitor and respond to changing conditions. Traffic data can be a useful tool for situational awareness due to its widespread availability. In many areas the use of traffic data can be easily utilized due to already existing infrastructure and low manpower requirements for collection and analysis.

Real-time trafﬁc data may be useful for a number of different func-tional needs in emergency operations. The statistic of how many vehi-cles or people are evacuating on a given route is useful for a number of reasons. This information could be used by transportation ofﬁcials to anticipate demand on other parts of the system. It can also provide information that will help transportation and law enforcement know

2500

Sunday Traffic 8/14

SR -1 @ Indian River

2000

Sunday Traffic 8/14

SR -1 @ DE 16

SR -1 @ Indian River

1500 500 1000

SB SB 0 2500 2000 1500 500 1000

0 NB NB 12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM 1200 1000 800 400 600

0 1200 1000 800 400 600

12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM 12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM 12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM 12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM 2500 2000 1500 500 1000

0

12:00 AM 6:00 AM 12:00 PM 6:00 PM 12:00 AM

Evacuation Traffic 8/26

SR -1 @ DE 16

Evacuation Traffic 8/26

_{SR -1 @ Indian River}

SB SB

NB NB

Labor Day Traffic 9/5

SR -1 @ Indian River

Labor Day Traffic 9/5

SR -1 @ DE 16

SB SB

NB

(8)

what places need traf_{fic control or other resources the most.} Emer-gency management officials concerned with receiving evacuees will be interested in this information so that they can plan for the needs of the incoming evacuees. These needs will depend on local evacua-tion practices. People may have informaevacua-tional needs such as what to do and where to go. The number of incoming evacuees can also be used to inform shelter management, so that they can adjust their staffing and logistics operations to appropriately anticipate housing evacuees.

The percentage of people which have evacuated and population remaining in an area is also a useful statistic. These statistics help transportation officials have an idea of how many people have evacu-ated and how many more people are expected to evacuate from an area. This information can also be used to adjust routing, make public messages and adjust traffic control as needed to get as many people out as possible and avoid leaving people on the road when the hurri-cane hits. The percent evacuated and population remaining can also help local public safety. Localfirst responders will be interested in knowing how much of the population is left so that they can be ready to respond after the hurricane to help those who have not evacuated.

5.2. Use of trafﬁc data for post-incident review and planning

Trafﬁc data from an evacuation can be very useful to ofﬁcials reviewing their actions in a hurricane and planning for the next

hurricane. The analysis of traf_{fic data can help emergency} manage-ment and transportation officials understand how the public and road network behave during a hurricane evacuation. This will allow them to make better plans. This will help them make connections that improve the way they manage traffic. They could also compare to new plans to the old traffic data to see if the plans are appropriate. 5.3. Observations related to evacuation behavior in Delaware

A number of observations can be made about the Delaware trafﬁc data analyzed from hurricane Irene.

Traffic data are useful to see how closely evacuation plans are followed, either in real-time or after the fact. For example, the com-munities of Fenwick Island, South Bethany and Bethany Beach are all instructed to evacuate west on DE 26first and then head north. This is somewhat similar to the path most travelers would take in normal circumstances. For example, the weekend before the evacua-tion 22% of vehicles leaving the area took DE 26 and 28% went north on DE 1 across the Indian River Inlet. The plans presumably direct traffic off of State Route 1 so that it does not become congested as it passes through Dewey, Rehoboth and Lewes.

Despite the plans to evacuate from Fenwick Island, South Bethany and Bethany Beach by going west on DE 26, as shown inFig. 6, and not north on DE 1, a large number of evacuees still went north. During the period of the evacuation, 30% of vehicles that exited the area used DE 26 and 38% left headed north on DE 1 crossing the Indian River inlet.

(9)

Despite evacuation plans, there was little difference in route selection be-tween DE 1 and DE 26. In both cases outbound trafﬁc on DE 26 was 79% of the volume for outbound trafﬁc on DE 1 at the Indian River inlet.

Even though the use of DE 1 was not according to plans, it did not appear to cause any problems. Twenty miles north on DE 1 the added demand from the Fenwick and Bethany area did not have a major im-pact when added to the trafﬁc from Dewey, Rehoboth and Lewes. Trafﬁc there peaked at 2253 vehicles per hour, which is only 63% of capacity.

5.4. Advantages/disadvantages of trafﬁc data usage

Trafﬁc data for evacuation monitoring and analysis have both ad-vantages and disadad-vantages. For real-time condition monitoring it can quickly provide some insight, but it requires more analysis to re-ally understand how many people evacuated from an area. Another problem is that the data do not distinguish among households evacu-ating to a safer location within town, to a nearby town or out of the area. For example, if some people are evacuating to a given town and others are evacuating from that town, it will be difﬁcult to tell how many people evacuated that town.

The usefulness of traffic data can also be limited by difficulties in analyzing data to provide useful information to decision-makers. There are many sources of this difficulty. Many times there will not be enough data to provide valuable information. This could be

because many traffic counters are located in urban areas and not the rural areas through which many people evacuate. Even when traf-fic counters are installed they may not function in an emergency due to loss of solar power or a broken data link. Of the 13 traffic counters looked at for this analysis two of them did not have data due to solar power loss during the weekend of the storm and one was setup incor-rectly making its data completely unusable.

Even when enough data are available it still takes time and effort to produce meaningful information to guide real-time response. Even with sufficient staffing it may be difficult to process the data from hundreds of devices and draw useful information for the dozens of locales affected by a hurricane. Fortunately, a number of computers systems exist to make this process easier during an evacuation

[25,38]. With practice and the right computer tools this can be done. For post disaster evaluation, trafﬁc data do not provide as much information as other methods, like phone interviews. Unlike phone interviews, trafﬁc data do not give much insight into who evacuated, where exactly they came from, where they went to or how they de-cided to evacuate. For this type of information written or internet sur-veys, and in person or phone interviews are still necessary.

6. Conclusions

Evacuation planning is a key element of disaster preparedness. For hurricanes, evacuation is one of the most effective and widely used

(10)

tactics to protect human life but evacuation is costly and disruptive. Furthermore, predicting the behavior of the population evacuating is challenging and there are rarely opportunities to test the plans prior to a signiﬁcant event. Hurricane Irene provided such an oppor-tunity. Coastal Delaware was placed under mandatory evacuation as Hurricane Irene approached the east coast of the United States in Au-gust 2011. This is peak season for the beach communities and many vacations were cut short.

Traffic data from automated traffic counters are used to explore traffic patterns before, during and after the event and make some in-ferences about the behavior of residents and visitors to the beach communities of coastal Delaware. The analysis suggests that in the case of Hurricane Irene in Delaware, the evacuation orders were ef-fective, provided sufficient lead time, and reached the intended audi-ence. These conclusions are supported by the analysis that shows that a signi_{ficant number of residents and visitors evacuated from the} beach communities. Furthermore, the traffic during the evacuation shows very similar patterns to the traffic patterns experienced on summer weekends, suggesting that there is adequate capacity for evacuation, but this traffic did not necessarily use the posted evacua-tion routes. Ideally survey work could be undertaken to reinforce the analysis.

These insights suggest that this type of analysis may also be of value for other events and in other communities where lead times permit a planned evacuation. When employed properly trafﬁc data can be useful for improving hurricane evacuation planning and mon-itoring an evacuation in real-time, and post event assessment of evac-uation plans.

Acknowledgments

This work is partially supported by the University of Delaware University Transportation Center. The trafﬁc data were provided by Delaware Department of Transportation. The authors appreciated Don Haas and Philip Petrucci's support for our many data requests. References

[1] J. Pielke, A. Roger, J. Gratz, C.W. Landsea, D. Collins, M.A. Saunders, R. Musulin, Normalized hurricane damage in the United States: 1900–2005, Natural Hazards Review 9 (2008) 29–42.

[2] E. Blake, E. Gibney, The Deadliest, Costliest, and Most Intense United States Trop-ical Cyclones from 1851 to 2010, In: NWS NHC-6, National Weather Service, 2011, p. 5.

[3] B. Wolshon, E. Urbina, C. Wilmot, M. Levitan, Review of policies and practices for hurricane evacuation. I: transportation planning, preparedness, and response, Natural Hazards Review 6 (2005) 129–142.

[4] R.C. National, Communicating Uncertainties in Weather and Climate Information: A Workshop Summary, The National Academies Press, 2003.

[5] R.W. Perry, M.R. Green, The role of ethnicity in the emergency decision-making process*, Sociological Inquiry 52 (1982) 306–334.

[6] P. Slovic, J.H. Flynn, M. Layman, Perceived risk, trust, and the politics of nuclear waste, Science 254 (1991) 1603–1607.

[7] P. Slovic, Perceived risk, trust, and democracy, Risk Analysis 13 (1993) 675–682. [8] D.E. Wenger, J.D. Dyke, T.D. Sebok, J.L. Neff, It's a matter of myths: an empirical exam-ination of individual insight into disaster response, In: in: M.D. Pugh (Ed.), Collective Behavior: A Source Book, West Publishing Co., New York, 1980, pp. 65–78.

[9] J.M. Nigg, Communication and behavior: organizational and individual response to warnings, In: in: R.R. Dynes, B. De Marchi, C. Pelanda (Eds.), Sociology of Disasters: Contribution of Sociology to Disaster Research, Franco Angeli Libri s.r.l, Milan, Italy, 1987, p. 103.

[10] E.L. Quarantelli, What should we study?: Questions and Suggestions for Re-searchers about the Concept of Disasters, Disaster Research Center, University of Delaware, Newark, Del, 1986.

[11] M. Perez-Lugo, The mass media and disaster awareness in Puerto Rico, Organiza-tion & Environment 14 (2001) 55–73.

[12] P.C. Stern, W.E. Easterling, National Research Council (U.S.), Panel on the Human Dimensions of Seasonal-to-Interannual Climate Variability, Making climate fore-casts matter, 1999.

[13] D.S. Mileti, Disasters by Design: A Reassessment of Natural Hazards in the United States, Joseph Henry Press, Washington, D.C., 1999.

[14] H.W. Fischer, Response to Disaster: Fact Versus Fiction & Its Perpetuation—The Sociology of Disaster, University Press of America, New York, 1994.

[15] K. Dow, S.L. Cutter, Emerging hurricane evacuation issues: Hurricane Floyd and South Carolina, Natural Hazards Review 3 (2002) 12–18.

[16] P. Edara, S. Sharma, C. McGhee, Development of a large-scale trafﬁc simulation model for hurricane evacuation—methodology and lessons learned, Natural Haz-ards Review 11 (2010) 127–139.

[17] B. Wolshon, B. McArdle, Trafﬁc impacts and dispersal patterns on secondary road-ways during regional evacuations, Natural Hazards Review 12 (2011) 19–27. [18] J. Kang, M.K. Lindell, C. Prater, Hurricane evacuation expectations and actual

be-havior in Hurricane Lili, Journal of Applied Social Psychology 37 (2007) 887–903. [19] B. Wolshon, B. McArdle, Temporospatial analysis of Hurricane Katrina regional evacuation trafﬁc patterns, Journal of Infrastructure Systems 15 (2009) 12–20. [20] B. Wolshon, Evacuation planning and engineering for Hurricane Katrina, The

Bridge, Spring 2006 (2006) 27.

[21] S.E. Chang, N. Nojima, Measuring post-disaster transportation system performance: the 1995 Kobe earthquake in comparative perspective, Transportation Research Part A: Policy and Practice 35 (2001) 475–494.

[22] H. Fu, C. Wilmot, Survival analysis-based dynamic travel demand models for hur-ricane evacuation, Transportation Research Record: Journal of the Transportation Research Board 1964 (2006) 211–218.

[23] B.M. Williams, A.P. Tagliaferri, S.S. Meinhold, J.E. Hummer, N.M. Rouphail, Simulation and analysis of freeway lane reversal for coastal hurricane evacuation, Journal of Urban Planning and Development 133 (2007) 61–72.

[24] M.K. Lindell, C.S. Prater, Critical behavioral assumptions in evacuation time esti-mate analysis for private vehicles: examples from hurricane research and plan-ning, Journal of Urban Planning and Development 133 (2007) 18–29. [25] FHWA, Assessment of State of the Practice and State of the Art in Evacuation

Transportation Management, FHWA–HOP-08-015, 2006.

[26] B. Wolshon, E.U. Hamilton, M. Levitan, C. Wilmot, Review of policies and practices for hurricane evacuation. II: trafﬁc operations, management, and control, Natural Hazards Review 6 (2005) 143–161.

[27] A. Ballard, N. Trout, D. Borchardt, K. Obeng-Boampong, B. Ullman, S. Venglar, A. Voigt, R. Rajbhandari, Hurricane Evacuation Trafﬁc Operations, Texas Transporta-tion Institute, 2008.

[28] Edwards, Kelcey, All Hazards Evacuation Annex: Transportation Incident and Event Management Plan, Kent County, Delaware, 2007.

[29] Edwards, Kelcey, All Hazards Evacuation Annex: Transportation Incident and Event Management Plan, New Castle County, Delaware, 2007.

[30] Edwards, Kelcey, All Hazards Evacuation Annex: Transportation Incident and Event Management Plan, Sussex County, Delaware, 2006.

[31] E. Baker, Behavioral Assumptions for Hurricane Evacuation Planning in the Delmarva Peninsula, U. S. Army Corps of Engineers Philadelphia District, 2003. [32] National Oceanographic and Atmospheric Administration, Severe weather events

archive, http://www.spc.noaa.gov/exper/archive/events/2012.

[33] L. Avila, J. Cangialosi, Tropical Cyclone Report: Hurricane Irene, National Hurri-cane Center, 2011. AL092011.

[34] Delaware, Governor Markell Expands Mandatory Evacuation Order to Include Residents in Areas Most at Risk, State of Delaware Newsroom, 2011. 2012. [35] Ocean City, Hurricane Irene Information, Ocean City, MD Blog, 2011. 2012. [36] DelDOT, Trafﬁc Book (2011), 2010.

[37] P. George, Gimme Shelter, In: Delaware Beach Life, 2011, p. 48, September 2011. [38] M. Hardy, K. Wunderlich, Evacuation Management Operations Modeling Assessment: